Daniel Lefevre

Oxidez

9 1 3

http://localhost

Oxidez

AI & ML interests

None yet

Recent Activity

new activity about 2 hours ago

unsloth/Qwen3.8-27B-GGUF:Extendable to 1000000, it says on the card.

repliedto savoji's post about 3 hours ago

A 284B-parameter DeepSeek-V4-Flash now runs on a 64 GB desktop with 0 GPUs: llama-modchip shrinks the checkpoint to 56.56 GiB (1.71 bpw) and decodes at 9.1 to 14.7 tok/s. The standard UD-IQ1_S build is 76.87 GiB, above the ~62.5 GiB usable on a 64 GB Linux box. The repo mixes precisions (routed experts mostly IQ1_S, MLA projections in a custom Fractal-64 type) and patches the CPU kernels. The model plus a 256K KV cache then sits at 57.61 to 58.9 GiB RSS with zero swap. The repo was created on 2026-10-29 and already has 744 stars and 470 points on Hacker News. 📊 The numbers • Size: 56.56 GiB, 20.31 GiB smaller than UD-IQ1_S • Prefill: 25.0 to 44.36 tok/s (+77%) on a 16-core Ryzen 9 9955HX • Decode: 5.91 tok/s unpatched, 9.06 to 9.95 tok/s at Top-6, 11.5 to 14.7 tok/s with the n-gram drafter ⚠️ Caveats • "Lossless" means bit-exact against the same re-quantized checkpoint. Versus UD-IQ1_S the author reports 0.9964 top-logit cosine similarity • The 98.5/100 quality score comes from the author's own gate, and the headline speeds come from one machine • The custom tensor type 40 is rejected by stock llama.cpp, Ollama and LM Studio 🛠️ Try it • Run scripts/check_cpu_readiness.py from the repo: it needs no model download • Run ./modchip bench for speed, then ./modchip eval to check the quality claims yourself Repo: https://github.com/synchopate/llama-modchip HN thread: https://news.ycombinator.com/item?id=50411961 Would you accept a custom runtime for 9 to 15 tok/s from a 284B model on a 64 GB box? 📚 Read it elsewhere • The deep dive on Medium: https://medium.com/p/e47ab3581f58 • The long read on Substack: https://sajjadpakdamansavoji.substack.com/p/llama-modchip-runs-a-284b-deepseek • The thread on X: https://x.com/saavoji/status/2119858773280141526 • Every version on sajjadpsavoji.github.io: https://sajjadpsavoji.github.io/blog#llama-modchip

new activity 9 days ago

XHToken/Spark-X2.6-4B-GGUF:A lot better for Hermes than Qwen 3.5 9B

View all activity

Organizations

None yet

New activity in unsloth/Qwen3.8-27B-GGUF about 2 hours ago

Extendable to 1000000, it says on the card.

#138 opened 7 days ago by

BigDeeper

replied to savoji's post about 3 hours ago

I have run DeepSeek-V4.1-Flash-1017-IQ2_XXS , on a Ryzen9 9900X with 128Gb DDR5 plus a RTX PRO 6000 Blackwell, but for agentic work, is slow. The decode speed i got: 20 to 25 tok/s , is not enough for serious work.
Optimal would be at least 45 to 60 tok/s . Qwen3.8-Flash-Next-IQ4_XS i get decode 120 to 130 tok/s and prompt processing around 6050 tok/s, much better for work. With context override at 1048576 (--override-kv qwen35.context_length=int:1048576 \ --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 262144 \ ) , is capable of doing some serious tasks, no problems at all.
I like DeepSeek model, but until they make a smaller and capable model, is not an option.

New activity in XHToken/Spark-X2.6-4B-GGUF 9 days ago

A lot better for Hermes than Qwen 3.5 9B

❤️ 2

#9 opened 9 days ago by

Arshaan786

New activity in unsloth/Qwen3.8-Flash-Next-GGUF about 1 month ago

Running IQ4_XS with MTP , crashing after a while.

➕ 1

#74 opened about 2 months ago by

Oxidez

New activity in FINAL-Bench/Darwin-180B-RSI about 1 month ago

How good this model is at agentic coding?

#1 opened about 1 month ago by

DedeProGames

New activity in incoai/Qwen3.8-27B-Splash about 2 months ago

DFLASH2 for Qwen3.8 Flash NExt

#10 opened about 2 months ago by

Oxidez

reacted to kostakoff's post with 👍 2 months ago

Post

3,186

Canceling My Pro Subscription

I'm officially canceling my Hugging Face Pro subscription today.
I supported this platform because it stood for true openness and neutrality. This acquisition by NVIDIA fundamentally changes that.

Here’s why I’m against this deal:
- Neutrality is dead. NVIDIA is a US-based company. This means US regulations will inevitably dictate platform policies, creating direct pressure on Chinese developers and anyone building open-weight models outside the US.
- Community over bureaucracy. NVIDIA is a massive, slow-moving corporation. This acquisition will likely drown the community in corporate processes and commercial interests. Soon, uploading a simple finetune might become a bureaucratic nightmare.
- Open vs. Proprietary. Hugging Face was built on open-source ideals. NVIDIA? They are a fiercely proprietary hardware company with a minimal track record of meaningful open-source contributions. They sell chips, not freedom.
- And to add insult to injury, NVIDIA has practically abandoned consumer RTX GPUs in 2026 to chase data center profits. Why would I pay them for "openness" when they've turned their back on the very developers who built this ecosystem?

I paid for openness. Not for a corporate takeover.

🤗 was about community.

12 replies

replied to NitrAI's post 2 months ago

It's always been like that. When a community handles to build something that becomes big, everyone wants to have a part of the cake. They start with a small piece and will end with the whole cake.